Neural Network Jet Flavour Tagging with the Upgraded ATLAS Inner Tracker Detector at the High-Luminosity LHC

As the High-Luminosity LHC (HL-LHC) era approaches, the most recent flavour tagging algorithms used in the ATLAS Collaboration are studied in order to characterize the full physics potential of the upgraded ATLAS detector and to get ready for the upcoming data-taking in 2029. The studies presented in this note exploit the most up-to-date simulation of the upgraded Inner Tracker (ITk) to assess the performance of the sophisticated flavour tagging algorithms recently developed for the early Run 3 data-taking using jets in $t\bar{t}$ and $Z^\prime$ events. The adaptation of algorithms based on deep sets or graph neural networks is in particular investigated for the first time. Large improvements are obtained with respect to the previous generation of flavour tagging algorithms studied in the context of the HL-LHC ATLAS upgrade.